kuilt-bolt/us.tractat.kuilt.bolt/DurabilityState/Degraded

Degraded

data class Degraded(val fromOffset: Long, val toOffset: Long, val reason: String, val cause: Throwable? = null) : DurabilityState

Frames in [fromOffset, toOffset) are written and readable, but not confirmed durable.

They are in the archive: whole, CRC-valid, and visible to every reader. What failed is the durability upgrade, which is why the appends that wrote them still reported AppendResult.Written — AppendResult.Failed means "the ops are lost" and invites a re-feed, which would write a second copy of a record already on disk.

Sticky, and it widens rather than resetting. This state persists until a later flush whose range covers the whole of [fromOffset, toOffset) succeeds; while failures continue the range grows to cover them all. That is what preserves a once-and-then-cleared EIO.

Constructors

Degraded

constructor(fromOffset: Long, toOffset: Long, reason: String, cause: Throwable? = null)

Properties

cause

val cause: Throwable?

the most recent underlying failure, when the backend had one to hand.

fromOffset

val fromOffset: Long

the first append offset whose durability is in doubt. A backend that flushes at page granularity reports the start of the page range it covered, which need not land on a frame boundary — so this may point inside an earlier frame. It over-claims the doubt rather than under-claiming it, which is the only direction that is safe here, and a consumer treating it as "everything from at least here" is reading it correctly.

reason

val reason: String

what the most recent failing flush said, errno text included where the backend has one. Earlier reasons are not accumulated; the range is what says how much is at risk.

toOffset

val toOffset: Long

one past the last such offset. May equal fromOffset: a flush that carried no frame of its own — a segment header — still leaves the archive's durability in doubt from that offset, and reporting no range at all would lose the fact entirely.